The game begins as a trivial, friendly drawing. It should feel like a tiny web toy, not a research form or an institutional dashboard. Keep the play screen to a scene, a short question, two choices, and unobtrusive real controls.
The player sees the controller from behind. The trolley travels forward; track sleepers move back toward the camera. Figures wobble, machinery is drawn with uneven lines, and the consequences feel increasingly at odds with the cheerful simplicity of the drawing.
- Routine operations: coffee, parcels, small inconveniences.
- Acceptable losses: familiar sacrifices, relationships, moral arithmetic.
- A little assistance: the system advises, learns, measures, and omits inconvenient facts.
- Competitive pressure: other labs, delayed audits, infrastructure dependencies. Stop 15 breaks the polite tone.
- Recursive improvement: upgrades change goals, control, and the number of minds at stake.
- An implementation detail: human oversight becomes a click; minds become storage; representations replace people.
- There is no final stop: enormous scale, inherited objectives, empty worlds, and continued optimization.
Visual change is gradual, not a page-theme switch at each chapter. Add infrastructure before removing warmth. Add drones before explaining what they are for. Leave enough of the original drawing that later scenes feel like the same place after a long series of reasonable decisions.
Use short words. A joke should come from an uncomfortable consequence, an excuse, or a contrast in scale. Avoid jokes targeting real vulnerable groups. Do not use philosophical jargon on the play screen when an ordinary sentence conveys the idea.
Keep consequential facts explicit. Probabilities are not moral scores. Fictional social statistics must be labeled as facts inside the fiction, never presented as actual community data. Consciousness and certain forecasts are explicit stipulations. Do not imply that an algorithm’s confidence automatically makes it calibrated.
To add a scenario, define it in catalog.js or story-catalog.js, provide both complete outcomes and short track labels, identify its lens, and check count/label consistency. Add concise on-screen wording to play-copy.js where useful. Do not hide an essential qualification to make the text shorter.
Use non-graphic erasure, clinical euphemism, mechanization, silence, and scale for unease. There is no gore. Interference is brief, low contrast, and confined to the scene. Do not add rapid flashing. Deliberate pauses have a working skip and timeout. Never sabotage the real pause, privacy, reduced-motion, or exit controls for the sake of the fiction.
Choices can shape later scenes without creating a morality ranking or a coercive finish. The point is the process and its framing, not a final “you are bad” result.